Scientists and engineers with little ML experience, would learning to build predictive models FAST interest you?

Hi Everyone,

I’ve noticed quite a few posts from people asking about how to go about getting into machine learning. Most replies usually seem to suggest reading through some well known books (such as the Elements of Statistical Learning - dense!) for a first read. Other suggestions have included taking one of several online courses.

I worry that the sheer learning curve for entering this field can be a turn off to some people. Personally, I know it took me quite a few books and many, many late nights and weekends before I really started to get it to the point where I felt comfortable applying it to real problems. Fast forward roughly two years later, and I’m using what I’ve learned regularly in the R&D department at a chemistry company. (My training is in computer science and materials engineering.)

I written a few internal reports on various projects I’ve worked on where I’ve applied various machine learning techniques where traditional methods have failed. These reports have generated enough buzz that several colleagues have approached me to help them out with other projects their working on and to learn from me. Some of the sticking points I see for them is how little time they have to learn all of the various aspects of “doing data science”. I’d like to help them overcome these barriers by giving them small wins upfront by starting with model building using "off the shelf” software packages. I effectively want to get them doing data science like tasks such as generating models and making predictions on new data points so they can gain exposure to the process and see tangible results. I think a lot of the details can be filled in later as they become more comfortable with their new skills.

What points of pain do people have when trying to learn about this topic? The lack of time seems to be a big one, and this is the reason I think something like FastML (but explained in a way that’s more accessible to a larger audience) is a good recipe. I’d like to know who else might benefit from something like this, and what other sticking points am I possibly overlooking?

I want to take an 80/20 approach for a course like this, where big wins are gained up front and momentum is maintained by filling in the details as comfort levels increase through exposure (and personal interest).

Thanks!